In message <[email protected]>,
Jonas Kahn <[email protected]> writes
< snip >
As far as I have understood, bridge programs would indeed use a MC
approach.
Typically, they will simulate 100 hands coherent with the information
they have. They then build the whole minimax tree with total information
(as if everybody knows everybody's hands) for each of these hands, and
choose the card with the best mean result, or that which gives the
contract most often. At the beginning, they might not build the whole
tree but play a few cards in the simulation using heuristics, but they
will build it when each player has ten cards.
If they had much more computing power, they would make simulations where
the players do not know the other hands (I think these are called
‹double dummy›), but that means that after the first card is played *in
"double dummy" means that all the players _do_ know all the hands.
the simulation*, the other simulated players have to make simulations
themselves... More or less the square of the number of simulations is
then required.
I think you can find a few explanations on Wbridge5's site, but I would
not swear it.
And here I've yielded all my knowledge.
You know more than I do.
Nick
--
Nick Wedd [email protected]
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
